- This is for multi-class short text classification.
- Model is built with Word Embedding, LSTM ( or GRU), and Fully-connected layer by Pytorch.
- A mini-batch is created by 0 padding and processed by using torch.nn.utils.rnn.PackedSequence.
- Cross-entropy Loss + Adam optimizer.
- Support pretrained word embedding (GloVe).
- Embedding --> Dropout --> LSTM(GRU) --> Dropout --> FC.
- The following command will download the dataset used in Learning to Classify Short and Sparse Text & Web with Hidden Topics from Large-scale Data Collections from here and process it for training.
- Also it download GloVe embeddings.
python preprocess.py
- The following command starts training. Run it with
-hfor optional arguments.
python main.py